Discrimination of Peanut Mildew Degree Based on Terahertz Attenuated Total Reflection Spectroscopy

In order to detect the different degrees of mildew of peanut kernels in an efficient, convenient and reliable way, a qualitative analysis method of mildew peanut based on back propagation(BP)neural network algorithm and support vector machine based on Terahertz (THz) time-domain spectroscopy was studied. In order to eliminate the contingency brought by different peanut samples, two peanut varieties, Huayu 36 and Luhua 9, were randomly collected for mildew culture. According to the sensory characteristics of peanut and the existing research foundation, the peanut samples were divided into four categories: normal, mild mildew, moderate mildew and severe mildew. The spectrum of peanut kernel samples (band 0.3~3.6 THz) was collected by THz total reflection. The Fourier transform method was used to perform frequency domain transformation on the time domain spectral signal and window processing. Then the optical constant absorbance and absorption coefficient of the obtained frequency domain signal were extracted, and the optical constant signal of the sample was obtained and the characteristic band was screened. On this basis, BP neural network qualitative analysis model and SVM qualitative analysis model were established respectively. Experiment results showed that the BP neural network model had a prediction set recognition rate of 88.57% for the Huayu 36 peanut mold model, and the prediction set recognition rate of the Luhua 9 peanut model was 91.40%; the Lib-SVM model for two varieties of peanut mold whether or not the two-class model, the three-class model of the three types of mildew peanuts had a prediction set recognition rate of 100%. It was shown that the application of Terahertz time-domain spectroscopy combined with BP neural network algorithm and SVM algorithm had a good effect on detecting mildewed peanut kernels. © 2019, Chinese Society of Agricultural Machinery. All right reserved.